Climate simulations by global climate models use rather coarse representations of the Earth, with grid-box sizes exceeding 100 km². These models assume that all the land, atmosphere and ocean features smaller than the grid boxes can be approximated as their averages over the grid areas. This assumption works quite well and the atmospheric and ocean circulation produced by the global models are generally very realistic.
However, we also know that local climate is strongly influenced by features such as mountains, coastline, lakes and so on, which cannot be described accurately on the global climate model grids. And impact studies and policy makers need climate scenarios including all the effects caused by local features. For this reason, techniques for downscaling global climate model simulations have been developed.
One of the most popular techniques is based on the use of regional climate models. These are usually limited area models (i.e., they are run over an area such as Europe) which include the atmospheric, land-surface and chemistry components similar to those in the global models. Since the simulated region is only a fraction of the globe (see diagram below), it is possible to increase the horizontal resolution to a few tenths of kilometre (25km² for the ENSEMBLES project European simulations). The increased spatial resolution is associated with a similar refinement in time, the typical regional climate model time-step is about five minutes while global models use about thirty minutes. These increased spatial and temporal resolutions allow the explicit description of many processes which have to be averaged in global climate model equations.
The regional climate models are driven by the atmospheric circulation at the boundary of the chosen region and the sea surface conditions taken from the global climate model simulations. The results are a set of climate variables for the chosen region which can be considered as high-resolution versions of the driving global model simulations. As in the case of the global models, these climate variables are consistent with the basic laws of physics, such as conservation of mass and energy and the Newtonian laws.
The other main approach is statistical downscaling which involves the application of relationships identified in the observed climate, between the large and smaller scale, to global climate model output. It assumes that the relationships bteween predictors (large-scale atmospheric circulation) and predictands (local weather) do not vary under climate change conditions. A number of user-friendly tools for statistical downscaling have been deveoped including the ENSEMBLES downscaling portal and the UKCP09 weather generator